How it works

Full guide for clients and GPU contributors.

For clients — step by step

You do not need Tailscale — only a browser. For jobs to run, the platform operator sets up Tailscale between the server and the GPU PC. Full guide: /tailscale Tailscale setup

  1. 1. Sign up as Client

    Open Sign up → choose Client. Sign in at gpugrid.duckdns.org.

  2. 2. Top up balance

    Dashboard → My jobs → Top up. Demo top-up adds USD to your account (no card yet).

  3. 3. Create a job

    New job → Title + Description. Put lyrics in Description (Verse/Chorus tags work). Set budget (USD). Submit.

  4. 4. Wait for the GPU

    Status pending → running → completed (usually 3–15 min for audio). Page refreshes every 8 seconds.

  5. 5. Get the result

    When status is completed, download MP3 via API: GET /api/tasks/{taskId}/artifact (Bearer token from Sign in).

For contributors — connect your GPU PC

  1. 1. Requirements

    Windows or Linux, NVIDIA GPU 8GB+ VRAM, Python 3.11–3.12, Node.js 18+, Tailscale on VPS and PC.

  2. 2. Install ACE-Step on your PC

    Clone github.com/ace-step/ACE-Step-1.5, run uv sync, set ACESTEP_API_KEY, start: uv run acestep-api (listens on :8001).

  3. 3. Tailscale

    Install Tailscale on PC and server, same account. Note your PC Tailscale IP (e.g. 100.x.x.x). Platform admin sets ACE_STEP_REMOTE_URL=http://YOUR_IP:8001 on the server.

  4. 4. Sign up & CLI

    Register as Contributor. Then:

    CONTRIBUTOR_API_URL=https://gpugrid.duckdns.org npx @contributor-ai/client login
CONTRIBUTOR_API_URL=https://gpugrid.duckdns.org npx @contributor-ai/client run
  5. 5. On the GPU machine

    Set ACE_STEP_REMOTE_URL=http://127.0.0.1:8001 and the same ACE_STEP_API_KEY before running the client. Worker runs ACE-Step for ace_step_audio tasks.

Windows quick start

  • Download contributor-setup.bat and contributor-run.bat from the contributor dashboard.
  • Run setup.bat once — enter contributor email/password.
  • Keep ACE-Step API running, then run contributor-run.bat.

FAQ

Why does the site say Remote GPU offline?
ACE-Step is not running on the PC, or Tailscale is disconnected. Start acestep-api and check Tailscale on both devices.
Do I need a contributor if the platform has remote GPU?
Jobs can run on the platform's connected PC automatically. Contributors are for extra capacity and other machines.
What task types are supported now?
Audio generation with lyrics via ACE-Step (ace_step_audio). More GPU task types coming later.